Aberrant Brain Network Dynamics in Childhood Autism and Its Relation to Behavioral Inflexibility

Thursday, May 12, 2016: 11:30 AM-1:30 PM
Hall A (Baltimore Convention Center)
K. Supekar, S. Ryali and V. Menon, Stanford University School of Medicine, Stanford, CA

There is a growing consensus that the underlying neurobiological disturbance associated with autism spectrum disorder (ASD) is aberrant large-scale brain networks. Remarkably, to date, there have been no systematic attempts to characterize dynamic/time-varying functional interactions among these networks in children with ASD, work that is critical for understanding the etiology of this complex neurodevelopmental disorder.  


To characterize dynamic functional interactions within a triple-network saliency model, and test the hypothesis that dynamic functional interactions among the salience (SN), central executive (CEN), and default mode (DMN) networks are dysregulated in children with ASD. We also determine whether dynamic network dysregulation measures can differentiate children with ASD from typically-developing (TD) children and predict clinical symptoms.


One cohort of 40 children (ASD=20,age:10.1±1.6;TD=20,age:10±1.6) and a second cohort of 60 children (ASD=30,age:10.3±1.8;TD=30,age:10±1.7) participated in the study. Task-free fMRI was acquired from both cohorts.

Regional fMRI timeseries were extracted from six key network-nodes: right anterior insula and anterior cingulate cortex (SN); right posterior parietal cortex and right dorsolateral prefrontal cortex (CEN); posterior cingulate cortex and medial prefrontal cortex (DMN). We developed a novel Hidden Markov (HMM)-based method that overcomes limitations of existing approaches for estimating dynamic/time-varying functional interactions between distributed brain regions. For each participant, we applied the HMM method to the regional timeseries to identify brain states and estimate probability of each brain state at each timepoint; each brain state is characterized by distinct intrinsic functional connectivity structure. The states and dwelling times of each state calculated as frequency of occurrence of that state across time, was compared between the two groups. To determine whether the learned HMM models can be used to discriminate children with ASD from TD children, we compared the likelihood of observing the TD data given the ASD model and vice-a-versa. Lastly, we measured the relationships between the state dwelling times and the ASD symptom severity.


In Cohort 1, in both groups we found two prominent brain states: ‘segregated state’ characterized by strong within-network coupling and no cross-network coupling, and ‘integrated state’ characterized by strong cross-network coupling. Notably, brain dynamics of ASD children were characterized by less frequent ‘integrated state’, than TD children. HMM-based classifiers differentiated children with ASD from TD children with 84% accuracy. ASD children who showed the lowest dwelling time in the ‘integrated state’ had the most severe restricted/repetitive behavior.

We repeated our entire analysis on the second cohort. In spite of differences in scanner and acquisition protocols, results from this analysis replicated our findings observed in Cohort 1.


Our findings provide not only provide novel evidence that alterations in dynamic coupling among the SN, CEN and DMN networks is a reproducible neurobiological signature of childhood autism but also show a link to core ASD symptomatology demonstrating for the first time that at earlier ages closer to disorder onset, the brain in children with ASD is inflexible in ways that contribute to behavioral inflexibility. More generally, the triple-network model provides a novel, replicable and parsimonious systems neuroscience framework for characterizing childhood ASD and predicting clinical symptoms in affected children.